8000 update MFA-Conformer performance after fixing the bug in #5797 by Jungjee · Pull Request #5826 · espnet/espnet · GitHub
[go: up one dir, main page]
More Web Proxy on the site http://driver.im/
Skip to content

update MFA-Conformer performance after fixing the bug in #5797 #5826

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Merged
merged 2 commits into from
Jul 12, 2024
Merged

update MFA-Conformer performance after fixing the bug in #5797 #5826

merged 2 commits into from
Jul 12, 2024

Conversation

Jungjee
Copy link
Contributor
@Jungjee Jungjee commented Jun 29, 2024

What?

There was a bug in the MFA-Conformer model's architecture which was fixed in #5797.
This PR updates the performance of the re-trained MFA-Conformer.
The model is available in Huggingface.

@sw005320 sw005320 added auto-merge Enable auto-merge SID Speaker identification/embedding labels Jun 29, 2024
@sw005320 sw005320 added this to the v.202405 milestone Jun 29, 2024
@codecov Codecov
Copy link
codecov bot commented Jun 29, 2024 •
edited
Loading

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 30.77%. Comparing base (e840295) to head (5655e26).
Report is 12 commits behind head on master.

❗ There is a different number of reports uploaded between BASE (e840295) and HEAD (5655e26). Click for more details.

HEAD has 1 upload less than BASE
Flag BASE (e840295) HEAD (5655e26)
test_integration_espnet1 1 0
Additional details and impacted files 
@@             Coverage Diff             @@
##           master    #5826       +/-   ##
===========================================
- Coverage   62.62%   30.77%   -31.86%     
===========================================
  Files         153      776      +623     
  Lines       17008    71485    +54477     
===========================================
+ Hits        10651    21997    +11346     
- Misses       6357    49488    +43131
Flag Coverage Δ
test_integration_espnet1 ?
test_python_espnet1 17.71% <ø> (?)
test_python_espnetez 13.97% <ø> (?)
test_utils 20.86% <ø> (?)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

@sw005320 sw005320 merged commit 639d3b5 into 8000 espnet:master Jul 12, 2024
34 checks passed
@sw005320
Copy link
Contributor

Thanks!

@Jungjee Jungjee deleted the spk_update_readme branch July 12, 2024 16:35
@series2
Copy link
Contributor
series2 commented Jul 15, 2024

@Jungjee
Thanks for the update the model.
I am currently trying to reproduce it.
However, I am having trouble because some of the published config are different from the config output when I experimented with conf/tuning/train_mfa_conformer_adamw.yaml.
Also, when I run the experiment with the latest state of the master branch, the eer.png file shows a tendency to overlearn at around 4 epochs, which was not seen before the fix.

Can you please release config:conf/tuning/train_mfa_conformer_adamw_fixEncoderBug2.yaml?
It would be great if you could release a reproducible environment, even if it is in the Jungjee/espnet repository.

Your kind response is always helpful.

@Jungjee
Copy link
Contributor Author
Jungjee commented Jul 15, 2024 •
edited
Loading

Hi @series2, sorry to give you the confusion.

Here are the updated parameters. Unstated parameters remain the same.

batch_size: 200
# Scheduler
scheduler: CosineAnnealingWarmupRestarts
scheduler_conf:
  first_cycle_steps: 300000
  cycle_mult: 1.0
  max_lr: 0.001
  min_lr: 0.00000001
  warmup_steps: 30000
  gamma: 0.7

Also please note that all parameters are available here.

8F20

@series2
Copy link
Contributor
series2 commented Jul 17, 2024

Thank you!

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers
No reviews
Assignees
No one assigned
Labels
auto-merge Enable auto-merge ESPnet2 README SID Speaker identification/embedding
Projects
None yet
Milestone
v.202409
Development

Successfully merging this pull request may close these issues.
3 participants
@Jungjee @sw005320 @series2
0